merge 'origin/sycl' into przemek/interop-map-memory

Author: DBDuncan

.github/docker/ubuntu-22.04.Dockerfile

@@ -32,7 +32,8 @@ ARG BASE_DEPS="\
 ARG UR_DEPS="\
 	doxygen \
 	python3 \
-	python3-pip"
+	python3-pip \
+	libhwloc-dev"
 
 # Unified Runtime's dependencies (installed via pip)
 ARG UR_PYTHON_DEPS="\

.github/scripts/install_hwloc.sh

@@ -0,0 +1,18 @@
+#!/usr/bin/env bash
+
+#  Copyright (C) 2024 Intel Corporation
+#  Part of the Unified-Runtime Project, under the Apache License v2.0 with LLVM Exceptions.
+#  See LICENSE.TXT
+#  SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+# install_hwloc.sh - Script for building and installing HWLOC library from source code
+
+set -e
+
+git clone -b hwloc-2.3.0 https://github.com/open-mpi/hwloc.git
+pushd hwloc
+./autogen.sh
+./configure
+make -j$(nproc)
+sudo make install -j$(nproc)
+popd

.github/workflows/build-hw-reusable.yml

@@ -11,6 +11,10 @@ on:
         required: false
         type: string
         default: ""
+      static_loader:
+        required: false
+        type: string
+        default: OFF
 
 permissions:
   contents: read
@@ -29,14 +33,19 @@ jobs:
     strategy:
       matrix:
         adapter: [
-          {name: "${{inputs.name}}", platform: "${{inputs.platform}}"},
+          {name: "${{inputs.name}}", platform: "${{inputs.platform}}", static_Loader: "${{inputs.static_loader}}"},
         ]
         build_type: [Debug, Release]
         compiler: [{c: gcc, cxx: g++}, {c: clang, cxx: clang++}]
         # TODO: The latest L0 loader segfaults when built with clang.
         exclude:
          - adapter: {name: L0, platform: ""}
            compiler: {c: clang, cxx: clang++}
+        # Exclude these configurations to avoid overloading the runners.
+         - adapter: {static_Loader: ON}
+           build_type: Release
+         - adapter: {static_Loader: ON}
+           compiler: {c: clang, cxx: clang++}
 
     runs-on: ${{matrix.adapter.name}}
 
@@ -63,6 +72,7 @@ jobs:
         -DUR_DEVELOPER_MODE=ON
         -DUR_BUILD_TESTS=ON
         -DUR_BUILD_ADAPTER_${{matrix.adapter.name}}=ON
+        -DUR_STATIC_LOADER=${{matrix.adapter.static_Loader}}
         -DUR_DPCXX=${{github.workspace}}/dpcpp_compiler/bin/clang++
         -DUR_SYCL_LIBRARY_DIR=${{github.workspace}}/dpcpp_compiler/lib
         ${{ matrix.adapter.name == 'HIP' && '-DUR_CONFORMANCE_AMD_ARCH=gfx1030' || '' }}

.github/workflows/cmake.yml

@@ -47,6 +47,9 @@ jobs:
         sudo apt-get update
         sudo apt-get install -y doxygen ${{matrix.compiler.c}}
 
+    - name: Install libhwloc
+      run: .github/scripts/install_hwloc.sh
+
     - name: Install g++-7
       if: matrix.compiler.cxx == 'g++-7'
       run: |
@@ -121,13 +124,20 @@ jobs:
     uses: ./.github/workflows/build-fuzz-reusable.yml
     with:
       test_label: "fuzz-short"
-    
+
   level-zero:
     name: Level Zero
     uses: ./.github/workflows/build-hw-reusable.yml
     with:
       name: L0
 
+  level-zero-static:
+    name: Level Zero static
+    uses: ./.github/workflows/build-hw-reusable.yml
+    with:
+      name: L0
+      static_loader: ON
+
   opencl:
     name: OpenCL
     uses: ./.github/workflows/build-hw-reusable.yml
@@ -217,11 +227,17 @@ jobs:
         Expand-Archive -Path "$WorkingDir\doxygen.zip"
         Add-Content $env:GITHUB_PATH "$WorkingDir\doxygen"
 
+    - name: Install hwloc
+      run: vcpkg install hwloc:x64-windows
+
     - name: Configure CMake
+      env:
+        VCPKG_PATH: "C:/vcpkg/packages/hwloc_x64-windows"
       run: >
         cmake
         -B${{github.workspace}}/build
         ${{matrix.toolset}}
+        -DCMAKE_PREFIX_PATH="${{env.VCPKG_PATH}}"
         -DCMAKE_C_COMPILER=${{matrix.compiler.c}}
         -DCMAKE_CXX_COMPILER=${{matrix.compiler.cxx}}
         -DCMAKE_POLICY_DEFAULT_CMP0094=NEW
@@ -260,6 +276,9 @@ jobs:
     - name: Install prerequisites
       run: python3 -m pip install -r third_party/requirements.txt
 
+    - name: Install hwloc
+      run: brew install hwloc
+
     - name: Configure CMake
       run: >
         cmake

.github/workflows/codeql.yml

@@ -31,6 +31,11 @@ jobs:
     - name: Install pip packages
       run: pip install -r third_party/requirements.txt
 
+    - name: Install apt packages
+      run: |
+        sudo apt-get update
+        sudo apt-get install -y libhwloc-dev
+
     - name: Configure CMake
       run: cmake -B ${{github.workspace}}/build -DUR_DEVELOPER_MODE=ON -DUR_BUILD_TESTS=ON -DUR_ENABLE_TRACING=ON -DUR_BUILD_TOOLS=ON -DUMF_ENABLE_POOL_TRACKING=ON
 
@@ -61,8 +66,13 @@ jobs:
     - name: Install pip packages
       run: python3 -m pip install -r third_party/requirements.txt
 
+    - name: Install hwloc
+      run: vcpkg install hwloc:x64-windows
+
     - name: Configure CMake
-      run: cmake -B ${{github.workspace}}/build -DCMAKE_POLICY_DEFAULT_CMP0094=NEW -DUR_DEVELOPER_MODE=ON -DUR_BUILD_TESTS=ON -DUR_ENABLE_TRACING=ON -DUR_BUILD_TOOLS=ON -DUMF_ENABLE_POOL_TRACKING=ON
+      env:
+        VCPKG_PATH: "C:/vcpkg/packages/hwloc_x64-windows"
+      run: cmake -B ${{github.workspace}}/build -DCMAKE_POLICY_DEFAULT_CMP0094=NEW -DUR_DEVELOPER_MODE=ON -DUR_BUILD_TESTS=ON -DUR_ENABLE_TRACING=ON -DUR_BUILD_TOOLS=ON -DUMF_ENABLE_POOL_TRACKING=ON -DCMAKE_PREFIX_PATH="${{env.VCPKG_PATH}}"
 
     - name: Build
       run: cmake --build ${{github.workspace}}/build -j $(nproc) --config Release

.github/workflows/coverity.yml

@@ -38,6 +38,11 @@ jobs:
       - name: Install pip packages
         run: pip install -r third_party/requirements.txt
 
+      - name: Install apt packages
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y libhwloc-dev
+
       - name: Configure CMake
         run: >
           cmake

.github/workflows/e2e_core.yml

@@ -65,6 +65,7 @@ jobs:
       with:
         files: |
           source/adapters/${{inputs.str_name}}/**
+          source/loader/**
 
   e2e-build-hw:
     # We want to run the job only if there are changes in the specific adapter

.github/workflows/e2e_level_zero.yml

@@ -21,9 +21,9 @@ jobs:
       config: ""
       unit: "gpu"
       # Failing tests
-      xfail: "ESIMD/regression/minmax.cpp;ESIMD/preemption.cpp;Matrix/SG32/element_wise_all_ops.cpp;Matrix/SG32/get_coord_int8_matB.cpp;Matrix/element_wise_all_ops.cpp;Matrix/element_wise_all_ops_1d.cpp;Matrix/element_wise_all_ops_1d_cont.cpp;Matrix/element_wise_all_ops_scalar.cpp;Matrix/element_wise_ops.cpp;Matrix/get_coord_int8_matB.cpp;Matrix/joint_matrix_apply_bf16.cpp;Matrix/joint_matrix_apply_two_matrices.cpp;Matrix/joint_matrix_bfloat16.cpp;Matrix/joint_matrix_bfloat16_array.cpp;Matrix/joint_matrix_rowmajorA_rowmajorB.cpp;ProgramManager/uneven_kernel_split.cpp"
+      xfail: "Matrix/SG32/get_coord_int8_matB.cpp;Matrix/element_wise_all_ops_1d.cpp;Matrix/element_wise_all_ops_1d_cont.cpp;Matrix/element_wise_all_ops_scalar.cpp;Matrix/get_coord_int8_matB.cpp;Matrix/joint_matrix_rowmajorA_rowmajorB.cpp;Plugin/level_zero_barrier_optimization.cpp"
       # Flaky tests
       filter_out: "UserDefinedReductions/user_defined_reductions.cpp"
       # These runners by default spawn upwards of 260 workers.
       # We also add a time out just in case some test hangs
-      extra_lit_flags: "-sv -j 100 --max-time 600"
+      extra_lit_flags: "--param gpu-intel-pvc=True -sv -j 100 --max-time 600"

.github/workflows/multi_device.yml

@@ -0,0 +1,63 @@
+---
+name: Multi Device testing
+
+on: [push, pull_request]
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
+  cancel-in-progress: true
+
+permissions:
+  contents: read
+
+jobs:
+  examples:
+    name: Multi Device testing
+    if: github.repository == 'oneapi-src/unified-runtime'  # run only on upstream; forks won't have the HW
+    strategy:
+      matrix:
+        adapter: [
+          {name: L0}
+        ]
+        build_type: [Debug, Release]
+        compiler: [{c: gcc, cxx: g++}] # TODO: investigate why memory-adapter-level_zero hangs with clang
+
+    runs-on: "${{matrix.adapter.name}}_2T"
+
+    steps:
+    - uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11 # v4.1.1
+
+    - name: Install pip packages
+      run: pip install -r third_party/requirements.txt
+
+    # TODO: enable once test failure are fixed/ignored
+    # - name: Download DPC++
+    #   run: |
+    #     wget -O ${{github.workspace}}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-01-29/sycl_linux.tar.gz
+    #     mkdir dpcpp_compiler
+    #     tar -xvf ${{github.workspace}}/dpcpp_compiler.tar.gz -C dpcpp_compiler
+
+    - name: Configure CMake
+      shell: bash -el {0}
+      run: >
+        cmake
+        -B${{github.workspace}}/build
+        -DCMAKE_C_COMPILER=${{matrix.compiler.c}}
+        -DCMAKE_CXX_COMPILER=${{matrix.compiler.cxx}}
+        -DCMAKE_BUILD_TYPE=${{matrix.build_type}}
+        -DUR_DEVELOPER_MODE=ON
+        -DUR_BUILD_TESTS=ON
+        -DUR_BUILD_ADAPTER_${{matrix.adapter.name}}=ON
+        -DUR_TEST_DEVICES_COUNT=2
+
+    - name: Build
+      run: cmake --build ${{github.workspace}}/build -j $(nproc)
+
+    - name: Test adapter specific
+      working-directory: ${{github.workspace}}/build
+      run: ctest -C ${{matrix.build_type}} --output-on-failure -L "adapter-specific" -E "test-adapter-level_zero_multi_queue" --timeout 180
+      # TODO: investigate why test-adapter-level_zero_multi_queue fails on newer driver
+
+    - name: Test adapters
+      working-directory: ${{github.workspace}}/build
+      run: env UR_CTS_ADAPTER_PLATFORM="${{matrix.adapter.platform}}" ctest -C ${{matrix.build_type}} --output-on-failure -L "conformance" --timeout 180

CMakeLists.txt

@@ -6,6 +6,11 @@
 cmake_minimum_required(VERSION 3.20.0 FATAL_ERROR)
 project(unified-runtime VERSION 0.10.0)
 
+# Check if unified runtime is built as a standalone project.
+if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR OR UR_STANDALONE_BUILD)
+    set(UR_STANDALONE_BUILD TRUE)
+endif()
+
 include(GNUInstallDirs)
 include(CheckCXXSourceCompiles)
 include(CMakePackageConfigHelpers)
@@ -37,7 +42,8 @@ option(UR_USE_MSAN "enable MemorySanitizer" OFF)
 option(UR_USE_TSAN "enable ThreadSanitizer" OFF)
 option(UR_ENABLE_TRACING "enable api tracing through xpti" OFF)
 option(UR_ENABLE_SANITIZER "enable device sanitizer" ON)
-option(UMF_BUILD_SHARED_LIBRARY "Build UMF as shared library" OFF)
+option(UR_ENABLE_SYMBOLIZER "enable symoblizer for sanitizer" OFF)
+option(UMF_BUILD_SHARED_LIBRARY "Build UMF as shared library" ON)
 option(UMF_ENABLE_POOL_TRACKING "Build UMF with pool tracking" ON)
 option(UR_BUILD_ADAPTER_L0 "Build the Level-Zero adapter" OFF)
 option(UR_BUILD_ADAPTER_OPENCL "Build the OpenCL adapter" OFF)
@@ -49,6 +55,7 @@ option(UR_BUILD_EXAMPLE_CODEGEN "Build the codegen example." OFF)
 option(VAL_USE_LIBBACKTRACE_BACKTRACE "enable libbacktrace validation backtrace for linux" OFF)
 option(UR_ENABLE_ASSERTIONS "Enable assertions for all build types" OFF)
 option(UR_BUILD_XPTI_LIBS "Build the XPTI libraries when tracing is enabled" ON)
+option(UR_STATIC_LOADER "Build loader as a static library" OFF)
 set(UR_DPCXX "" CACHE FILEPATH "Path of the DPC++ compiler executable")
 set(UR_DPCXX_BUILD_FLAGS "" CACHE STRING "Build flags to pass to DPC++ when compiling device programs")
 set(UR_SYCL_LIBRARY_DIR "" CACHE PATH
@@ -134,6 +141,15 @@ if(UR_ENABLE_TRACING)
         set_target_properties(xptifw PROPERTIES
             LIBRARY_OUTPUT_DIRECTORY ${CMAKE_LIBRARY_OUTPUT_DIRECTORY}
         )
+
+        if (UR_STATIC_LOADER)
+            install(TARGETS xpti xptifw
+                EXPORT ${PROJECT_NAME}-targets
+                ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}
+                RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR}
+                LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}
+            )
+        endif()
     endif()
 
     if (MSVC)
@@ -153,6 +169,14 @@ if(UR_ENABLE_SANITIZER)
     else()
         add_compile_definitions(UR_ENABLE_SANITIZER)
     endif()
+
+    if(UR_ENABLE_SYMBOLIZER AND UR_STANDALONE_BUILD)
+        find_package(LLVM REQUIRED)
+    endif()
+else()
+    if(UR_ENABLE_SYMBOLIZER)
+        message(FATAL_ERROR "Symbolizer must be enabled with Sanitizer layer")
+    endif()
 endif()
 
 if(UR_USE_ASAN)
@@ -179,7 +203,7 @@ endif()
 if(UR_FORMAT_CPP_STYLE)
     find_program(CLANG_FORMAT NAMES clang-format-15 clang-format-15.0 clang-format)
 
-    if(CLANG_FORMAT)        
+    if(CLANG_FORMAT)
         get_program_version_major_minor(${CLANG_FORMAT} CLANG_FORMAT_VERSION)
         message(STATUS "Found clang-format: ${CLANG_FORMAT} (version: ${CLANG_FORMAT_VERSION})")